Representing a generalizations distribution table by connectionist networks for evolutionary rule discovery

Abstract

This paper introduces a new approach for rule discovery from databases, in which a variation of transition matrix named generalizations distribution table (GDT) is used as a hypothesis search space for generalization. Furthermore, by representing the GDT as connectionist networks, if-then rules can be discovered in an evolutionary, parallel-distributed cooperative mode. The key features of this approach are that it can predict unseen instances because the search space considers all possible combination of the seen instances, and the uncertainty of a rule including the prediction of possible instances can be explicitly represented in the strength of the rule. This paper focuses on some basic concepts of our methodology and how to represent generalizations distribution tables by connectionist networks.